Introducing Gemini 3.8 Flash and 3.8 Flash Cyber

Google has released Gemini 3.8, its newest and best reasoning and coding model, offering two variants: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. Gemini 3.8 Flash is a highly capable "workhorse" model with significant improvements over the previous 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning, while maintaining the same low cost and speed. It demonstrates performance approaching larger, more expensive frontier models on benchmarks like DeepSWE for long-horizon software engineering and specialized financial and legal agent tasks. A core design choice is that the model "works harder" on complex problems by executing more reasoning steps, with developers able to adjust effort levels to balance performance and efficiency. Gemini 3.8 Flash Cyber is a specialized model focused on autonomous vulnerability discovery and automated patching, achieving frontier-level performance. It surpasses previous models and larger competitors on benchmarks like CyberGym for vulnerability finding and CWE-Bench for patching, offering a high success rate at a lower cost. Early real-world use within Google shows significant improvements, such as the Chrome Security team finding it produced 2.6 times more correct patches for Chrome vulnerabilities than larger commercial models, and the Cloud Vulnerability Research team using it to discover a critical foundational vulnerability in under two hours. Both models are built with safety safeguards, with the Cyber variant available only to trusted defenders through a new "Fairwind Program."

(Source：Gemini)